Login

Publications  •  Project Statistics

Glossary  •  Schools  •  Disciplines
People Search: 
   
Title/Abstract Search: 

Dissertation Information for Hongyan Ma

NAME:
- Hongyan Ma

DEGREE:
- Ph.D.

DISCIPLINE:
- Library and Information Science

SCHOOL:
- University of California, Los Angeles (USA) (2008)

ADVISORS:
- Gregory H. Leazer
- Jonathan Furner

COMMITTEE MEMBERS:
- Christine L. Borgman
- Junghoo Cho
- Mark H. Hansen

MPACT Status: Fully Complete

Title: User-system coordination in Unified Probabilistic Retrieval: Exploiting search logs to construct common ground

Abstract: With the increasing use of Web search engines, there evolve acute needs for more adaptive and more personalizable Information Retrieval (IR) systems. Search logs have been investigated to gather useful data about contexts and users to support adaptive retrieval, but few studies have been undertaken that either construct a model of IR that provides a formal justification for the use of search logs, or carry out an empirical test of the retrieval effectiveness and real users' appeal of systems that exploit search-log data. This study proposes an IR Coordination Model, which applies Clark's coordination theory and ideas developed in IR cognitive models to a unified version of two probabilistic models of IR. This new model provides a motivation for the dynamic collection and exploitation of three kinds of data - linguistic, perceptual, and community membership to construct the "common ground" between IR mechanisms and searchers - data that can be collected via search logs.

This study tests the operation of a system, the UPIR (Unified Probabilistic Information Retrieval) system, which was designed on the basis of the IR Coordination Model and exploited data about common ground. Real users' search logs from two operational Web search engines, Infocious and Excite, were processed to obtain implicit feedback for adaptive indexing and query expansion. A TREC-setting experiment and a real user study were conducted to examine how users' relevance feedback and incremental community membership information influence search results, and how users' perceptual evidence on search results and searching process correlates with system performance.

The results demonstrate that the log-based UPIR system yields statistically superior performance over the baseline system. Search logs are useful data sources of linguistic, perceptual, and community membership information. The results show very clearly that some perceptual differentials on search results and searching process correlate positively with retrieval performance. Accumulated community information based on queries and click-through data in search logs improves search performance but at a diminishing rate. This study thus far suggests that linguistic, perceptual, and community membership information are vital in constructing common ground, improving user-system coordination in IR interaction, and supporting adaptive retrieval.

MPACT Scores for Hongyan Ma

A = 0
C = 0
A+C = 0
T = 0
G = 0
W = 0
TD = 0
TA = 0
calculated 2009-05-25 19:43:49

Advisors and Advisees Graph